It is possible to understand some models (particularly popular statistical models) by learning how to interpret their coefficients. We’re not going to take that approach here for two reasons:
It requires many months of study to interpret the coefficients of even fairly simple model classes like logistic regression or survival analysis.
Modern model classes (like random forests or SVMs) may have 100s or 1000s (or more!) coefficients.
Instead, we’ll understand a model by looking at it’s predictions. This doesn’t cover every possible type of model (because not every model makes predictions) but it does cover a very wide class, including most models in statistics and machine learning.